In three-dimensional video encoding or multi-visual angle video encoding, there is a video with multi-visual angle which needs to be encoded. In order to make full use of the correlation between videos with adjacent visual angles, the existing method not only uses an image in an encoded visual angle to predict an image in a visual angle to be encoded, but also uses a residual image in an encoded visual angle to predict a residual image in a visual angle to be encode.
On one hand, since an inter-visual predicted residual is not highly relevant in different visual angles, the inter-visual predicted residual is not used to predict any residual; on the other hand, a time domain predicted residual is used to predict the time domain predicted residual and the inter-visual predicted residual. However, due to the different characteristics between the time domain predicted residual and the inter-visual predicted residual, predicting the inter-visual predicted residual by using the time domain predicted residual will often cause residual energy to be increased, thereby reducing the encoding efficiency.